What underlying language models does Burna AI use?

Burna AI uses frontier language models from major providers, currently centered on the latest GPT family for the default and medical models, with the model identifier surfaced in the audit trail for every grade. Model selection is encapsulated behind a model provider abstraction so the engine can evolve as frontier capabilities improve. For Protocol Safe deployments, attribution reasoning can run on a self-hosted MedGemma model inside the sponsor's own cloud, with BAAI/bge-large-en-v1.5 embeddings for retrieval. The architecture is model-aware but not model-locked: cascading constraints, citation enforcement, and per-drug attribution scoring are properties of the pipeline, not of any one model.